Kaprin
All insightsFDE Field Notes

How an FDE Engagement Should End, and What You Get at Handoff

By Kaprin Engineering 7 min read

This post is for buyers of forward deployed engineering (FDE), where a senior engineer works inside your systems and your team, and for the CTOs who will inherit the result. The worry is a fair one. You bring in an expert, the system works, the expert leaves, and nobody on your side can touch it.

The way around that is to treat the ending as a deliverable. Design it in week one, build it throughout, and test it before the engagement closes.

Why should an FDE engagement end at all?

An embedded engineer exists to get a system into production and transfer the capability to your team. Forrester describes FDEs as training wheels: useful, deliberately temporary, and meant to come off. A vendor whose footprint never shrinks is selling dependence.

In its analysis of the wave of FDE investment, Forrester argues that services firms should sell outcomes, build for handoff, and let their footprint shrink. We agree, and it shapes how we plan every engagement. Handoff by day 45 is one of four fixed beats in our rhythm, next to Scope, Embed and Harden.

There's a commercial honesty to it too. If the only way to keep a system running is to keep us, the system wasn't finished.

What artifacts make up a proper handoff?

A proper handoff has six artifacts: runbooks, architecture decision records, an evaluation suite, dashboards, access and credential ownership in your account, and a record of changes your own engineers shipped. If any one is missing, the handoff is incomplete.

ArtifactWhat it containsThe test of whether it works
Runbooks How to deploy, roll back, rotate keys, replay failed jobs, and respond to each alert. An engineer who didn't build the system follows one during a drill.
Architecture decision records Each significant choice, the alternatives considered, and the reason. Short, dated, in the repository. A new hire can answer "why is it built this way" without asking us.
Evaluation suite Test cases and checks that run before any change reaches production, including past failures turned into tests. Your team can change a prompt, a rule or a dependency and see what broke.
Dashboards and alerts Throughput, error rates, spend, latency, queue depth, and the business metric agreed at the start. Your owner can see a problem before a customer reports it.
Access and credentials Cloud account, repository, secrets manager, model provider keys and admin roles, all in your organization. We could lose access tomorrow and nothing would stop.
Shipped changes by your engineers Pull requests, reviews and deployments done by your team during the engagement. Your engineers have already done it once, in production, with us watching.

Why does paired working matter more than documents?

Documents describe a system, and paired working proves your engineers can change it. We pair with your team on real tickets in the later weeks. By handoff they've reviewed our code, shipped their own, and deployed to production at least once while we're still around to catch mistakes.

Documentation alone fails in a predictable way. It's accurate on the day it's written, then it drifts. An engineer who has made a change, watched it break in staging and fixed it understands the system in a way no runbook can give them.

This is also why the code lives in your account from the start. At Kaprin the client owns 100% of the code, database and IP, deployed in your own cloud account (AWS or GCP) or private VPC. There's no export step at the end because nothing was ever held back.

What are the exit criteria?

Exit criteria are the written conditions that say the engagement is finished. Agree them in week one, when nobody is tempted to bend them. A good set covers the metric, the stability of the system, the handoff pack and your team's ability to operate it alone.

  1. The metric has moved. The baseline agreed in week one has improved by the amount you decided counts as success, measured on live traffic.
  2. The system is stable in production. It has run for an agreed period without a severity-one incident that needed us.
  3. All six artifacts above exist, sit in your repository or account, and have been reviewed by your owner.
  4. One person on your side owns the system and its backlog, and has the access to act.
  5. Your engineers have shipped at least one change, from ticket to production, with nobody from us on the keyboard.
  6. You've passed a drill: rolled back a deployment and replayed a failed job using only the runbooks.

If one of these isn't met, we say so and extend the sprint by a defined amount. We don't quietly declare victory.

What does a post-launch support retainer include?

A retainer covers a defined set of recurring work on the system we delivered: reviewing dashboards and alerts, fixing defects in the delivered scope, upgrading dependencies and model integrations, and keeping the eval suite current. It exists to keep a handoff safe. It shouldn't hold the system hostage.

The lists below show how we think a retainer should be framed. The specific terms for your engagement are agreed in writing.

Typically included

  • Periodic review of dashboards, alerts and spend against agreed thresholds.
  • Bug fixes for behavior that was in scope at handoff.
  • Dependency, security-patch and model-provider changes that would otherwise break the system.
  • Updates to the eval suite when production surfaces a new failure.
  • Answers to your engineers' questions, and reviews of their changes if they ask.

Typically not included

  • New workflows or major features. Those are a new scoped sprint.
  • Acting as your primary on-call. Your owner holds the pager unless you agree otherwise in writing.
  • Holding credentials, deployment rights or knowledge that your team doesn't also have.
  • Open-ended availability as a substitute for a team of your own.

How do you spot a vendor that is not building for handoff?

Watch for four signals: credentials that live in the vendor's accounts, no documentation until the last week, work done in private branches, and no request for a named owner on your side. Any one of these suggests the system will be hard to run without them.

The independent-versus-vendor question matters here too. CIO.com has reported that analysts such as Info-Tech and Carmi Levy warn about vendor lock-in with hyperscaler FDEs, in its report on Microsoft and Amazon's FDE investment. Whoever you hire, ask where the keys sit.

How does this fit into the full engagement?

Handoff work starts long before the last week. Runbooks and decision records begin in week two, the eval suite grows through hardening, and pairing begins as soon as the first workflow is stable. The week-by-week view of the first 90 days shows where each artifact appears.

For the technical side, see how we harden a prototype for production. If the term is new to you, our overview of what forward deployed engineering is is a good place to start.

What does a good handoff day look like?

On a good handoff day nothing dramatic happens, and that's how you know it worked. Your owner opens the dashboards, your engineer deploys a small change, someone runs the rollback drill from the runbook, and we confirm that every credential and every repository sits in your accounts. Then we close the checklist together.

If the day feels tense, the handoff started too late. By then we should mostly be watching, answering questions and signing off, because your team has been doing the work for weeks. We also agree what happens next: a light retainer, a second sprint on a new workflow, or nothing at all.

Next step

If you'd like to see what a handoff pack would look like for your own system, get in touch. We can walk through a sample outline and the exit criteria we'd propose.

Questions readers ask

What should an FDE hand over at the end of an engagement?

At minimum, the code and database in your own cloud account, runbooks, architecture decision records, an evaluation suite, dashboards, and ownership of every credential. Your engineers should also have shipped changes to the system alongside the vendor.

Why would an FDE firm want to leave?

A system you can't run without us is a liability for you and a distraction for us. Forrester advises services firms to build for handoff and let their footprint shrink. A good engagement ends with a smaller footprint and a team that can carry on.

How do I avoid vendor lock-in with an FDE?

Make sure the code, database and cloud account belong to you from day one, that credentials sit in your accounts, and that your engineers pair on real changes. Lock-in usually comes from access and knowledge the vendor holds, and much less from the code itself.

Does a post-launch support retainer mean the engagement never ends?

It shouldn't. A retainer covers a defined set of work, such as monitoring review, bug fixes within the delivered scope and dependency upgrades. It doesn't replace your owner, and new features belong in a new scoped sprint.

What are good exit criteria for an FDE engagement?

Agree them in week one. Typical ones are that the baseline metric has moved, the system has run in production for an agreed period, the handoff pack is complete, and your engineers have shipped at least one change without help. Sign-off happens against that list.

Bring us the broken prototype, the SaaS bill, or the manual process

Tell us what is stuck. We reply within 24 hours with a scoped plan, a fixed timeline and the one number we will move.

More insights

FDE = forward deployed engineering. Strict NDA protection. Senior architect review.